Our approach is easy to deploy incrementally in existing legacy NFS environments. So long as the same underlying file system (and hence, disk layout) is kept, we could take existing disks with data and use them (without reformatting) as NADs by attaching them to the network via a controller that checks for capabilities and replay attacks. (Of course, this controller would have to be specially manufactured for use with our protocol.) By having our metadata server also export the Snapdragon file system via plain NFS, we can support legacy NFS clients, albeit with poorer performance. This allows an NFS system to be converted one client at a time.